source('Core_functions.R')
source('Simulation_functions.R')
source('plotting_functions.R')
source('wrapper_function_all_scenarios.R')
TEL = 0.95 # Target Efficacy Level
MTT = 0.05 # Maximum Tolerated Toxicity
N_max = 250 # Maximum number of patients recruited
max_increment = 1 # Maximum dose increment to doses previously unseen
N_trials = 2000
N_batch = 3
Randomisation_p_SOC = 0.2 # proportion randomised to the standard of care dose
starting_dose = 12 # starting dose in adaptive arm
SoC = 8 # Standard of Care
# function that solves for the beta value based on interpretable parameters
solve_beta = function(alpha_val, v_star, y_star){
beta_val = ( logit(y_star) - alpha_val ) / log2(v_star)
return(beta_val)
}
#******* Prior point estimates *********
Prior_TED = 12; # prior estimate of the Target Efficacious Dose
Prior_alpha_eff = logit(1/10) # prior estimate of the efficacy with one vial
Prior_beta_eff = solve_beta(alpha_val = Prior_alpha_eff, v_star = Prior_TED, y_star = TEL)
Prior_MTD = 32; # prior estimate of the Maximum Tolerated Dose
Prior_alpha_tox = logit(1/1000) # prior estimate of the toxicity after 1 vial
Prior_beta_tox = solve_beta(alpha_val = Prior_alpha_tox, v_star = Prior_MTD, y_star = MTT)
#******* Prior uncertainty estimates *******
prior_model_params = list(beta_tox = Prior_beta_tox,
beta_tox_sd = .05,
alpha_tox=Prior_alpha_tox,
alpha_tox_sd = 2,
beta_eff=Prior_beta_eff,
beta_eff_sd = .05,
alpha_eff=Prior_alpha_eff,
alpha_eff_sd = 2)
This function computes linear dose-response curves based on the parameters (MTT, TEL etc).
make_linear_dose_response = function(MTT, true_MTD, tox_zero=1,
TEL, true_TED, eff_zero=1){
log_xs = log2(seq(0, 10000, by=.1))
tox_MTD = true_MTD
slope_tox = MTT/(log2(tox_MTD) - log2(tox_zero))
intercept_tox = -slope_tox * log2(tox_zero)
ys = slope_tox*log_xs + intercept_tox
ys[ys>1]=1; ys[ys<0]=0
f_true_tox = approxfun(x = 2^log_xs, y = ys, rule = 2)
eff_TED = true_TED
slope_eff = TEL/(log2(eff_TED) - log2(eff_zero))
intercept_eff = -slope_eff * log2(eff_zero)
ys = slope_eff*log_xs + intercept_eff
ys[ys>1]=1; ys[ys<0]=0
f_true_eff = approxfun(x = 2^log_xs, ys, rule = 2)
true_model = list(tox = f_true_tox, eff = f_true_eff)
return(true_model)
}
true_alpha_eff = logit(1/50)
true_TED = 20 #
true_beta_eff = solve_beta(alpha_val = true_alpha_eff, v_star = true_TED, y_star = TEL)
true_alpha_tox = logit(1/500) # toxicity at 1 vial
true_MTD = 8 # simulation truth for the MTD
true_beta_tox = solve_beta(alpha_val = true_alpha_tox, v_star = true_MTD, y_star = MTT)
# well-specified simulation truth
model_params_true = list(beta_tox = true_beta_tox,
alpha_tox=true_alpha_tox,
beta_eff=true_beta_eff,
alpha_eff=true_alpha_eff)
# mis-specified simulation truth
true_model = make_linear_dose_response(MTT, true_MTD, tox_zero=1,
TEL, true_TED, eff_zero=1)
run_all_scenarios(model_params_true = model_params_true,
true_model = true_model,
prior_model_params = prior_model_params,
N_trials = N_trials,
MTT = MTT, TEL = TEL,
N_max = N_max,
N_batch = N_batch,
max_increment = max_increment,
Randomisation_p_SOC = Randomisation_p_SOC,
sim_title = 'Simulation scenario 1',
FORCE_RERUN=FORCE_RERUN,
N_cores = N_cores,
starting_dose = starting_dose,
SoC = SoC, use_SoC_data = T)
## Simulation scenario 1 , model based design, well-specified, ...
## [1] "Simulation scenario 1_model_based_well_specified_All_data.RData"
## 0.068 sec elapsed
## Simulation scenario 1 , model based design, mis-specified, ...
## [1] "Simulation scenario 1_model_based_mis_specified_All_data.RData"
## 0.051 sec elapsed
## Simulation scenario 1 , rule based design, well-specified, ...
## [1] "Simulation scenario 1_rule_based_well_specified_All_data.RData"
## 0.215 sec elapsed
## Simulation scenario 1 , rule based design, mis-specified, ...
## [1] "Simulation scenario 1_rule_based_mis_specified_All_data.RData"
## 0.176 sec elapsed
# Comparison in well-specified case
compare_rule_vs_model(sim_title = 'Simulation scenario 1',
model_params_true = model_params_true,
true_model = NULL,
prior_model_params = prior_model_params,use_SoC_data = T)
## For the rule-based design, 12% of trials give patient 252 a dose within +/-10% of the true optimal dose
## For the model-based design, 26% of trials give patient 252 a dose within +/-10% of the true optimal dose
# Comparison in mis-specified case
compare_rule_vs_model(sim_title = 'Simulation scenario 1',
model_params_true = NULL,
true_model = true_model,
prior_model_params = prior_model_params,use_SoC_data = T)
## For the rule-based design, 2% of trials give patient 252 a dose within +/-10% of the true optimal dose
## For the model-based design, 10% of trials give patient 252 a dose within +/-10% of the true optimal dose
run_all_scenarios(model_params_true = model_params_true,
true_model = true_model,
prior_model_params = prior_model_params,
N_trials = N_trials,
MTT = MTT, TEL = TEL,
N_max = N_max,
N_batch = N_batch,
max_increment = max_increment,
Randomisation_p_SOC = Randomisation_p_SOC,
sim_title = 'Simulation scenario 1',
FORCE_RERUN=FORCE_RERUN,
N_cores = N_cores,
starting_dose = starting_dose,
SoC = SoC, use_SoC_data = F)
## Simulation scenario 1 , model based design, well-specified, ...
## [1] "Simulation scenario 1_model_based_well_specified_Adaptive_data.RData"
## 0.06 sec elapsed
## Simulation scenario 1 , model based design, mis-specified, ...
## [1] "Simulation scenario 1_model_based_mis_specified_Adaptive_data.RData"
## 0.052 sec elapsed
## Simulation scenario 1 , rule based design, well-specified, ...
## [1] "Simulation scenario 1_rule_based_well_specified_All_data.RData"
## 0.179 sec elapsed
## Simulation scenario 1 , rule based design, mis-specified, ...
## [1] "Simulation scenario 1_rule_based_mis_specified_All_data.RData"
## 0.102 sec elapsed
# Comparison in well-specified case
compare_rule_vs_model(sim_title = 'Simulation scenario 1',
model_params_true = model_params_true,
true_model = NULL,
prior_model_params = prior_model_params,use_SoC_data = F)
## For the rule-based design, 12% of trials give patient 252 a dose within +/-10% of the true optimal dose
## For the model-based design, 25% of trials give patient 252 a dose within +/-10% of the true optimal dose
# Comparison in mis-specified case
compare_rule_vs_model(sim_title = 'Simulation scenario 1',
model_params_true = NULL,
true_model = true_model,
prior_model_params = prior_model_params,use_SoC_data = F)
## For the rule-based design, 2% of trials give patient 252 a dose within +/-10% of the true optimal dose
## For the model-based design, 8% of trials give patient 252 a dose within +/-10% of the true optimal dose
true_alpha_eff = logit(1/20)
true_TED = 8 # simulation truth for the MED
true_beta_eff = solve_beta(alpha_val = true_alpha_eff, v_star = true_TED, y_star = TEL)
true_alpha_tox = logit(1/500) # toxicity at 1 vial
true_MTD = 20 # simulation truth for the MTD
true_beta_tox = solve_beta(alpha_val = true_alpha_tox, v_star = true_MTD, y_star = MTT)
# well-specified simulation truth
model_params_true = list(beta_tox = true_beta_tox,
alpha_tox = true_alpha_tox,
beta_eff = true_beta_eff,
alpha_eff = true_alpha_eff)
# mis-specified simulation truth
true_model = make_linear_dose_response(MTT, true_MTD, tox_zero=1,
TEL, true_TED, eff_zero=1)
run_all_scenarios(model_params_true = model_params_true,
true_model = true_model,
prior_model_params = prior_model_params,
N_trials = N_trials,
MTT = MTT, TEL = TEL,
N_max = N_max,
N_batch = N_batch,
max_increment = max_increment,
Randomisation_p_SOC = Randomisation_p_SOC,
sim_title = 'Simulation scenario 2',
FORCE_RERUN=FORCE_RERUN,
N_cores = N_cores,
starting_dose = starting_dose,
SoC = SoC,
use_SoC_data = T)
## Simulation scenario 2 , model based design, well-specified, ...
## [1] "Simulation scenario 2_model_based_well_specified_All_data.RData"
## 0.054 sec elapsed
## Simulation scenario 2 , model based design, mis-specified, ...
## [1] "Simulation scenario 2_model_based_mis_specified_All_data.RData"
## 0.058 sec elapsed
## Simulation scenario 2 , rule based design, well-specified, ...
## [1] "Simulation scenario 2_rule_based_well_specified_All_data.RData"
## 0.116 sec elapsed
## Simulation scenario 2 , rule based design, mis-specified, ...
## [1] "Simulation scenario 2_rule_based_mis_specified_All_data.RData"
## 0.105 sec elapsed
# Comparison in well-specified case
compare_rule_vs_model(sim_title = 'Simulation scenario 2',
model_params_true = model_params_true,
true_model = NULL,
prior_model_params = prior_model_params,use_SoC_data = T)
## For the rule-based design, 32% of trials give patient 252 a dose within +/-10% of the true optimal dose
## For the model-based design, 43% of trials give patient 252 a dose within +/-10% of the true optimal dose
# Comparison in mis-specified case
compare_rule_vs_model(sim_title = 'Simulation scenario 2',
model_params_true = NULL,
true_model = true_model,
prior_model_params = prior_model_params,use_SoC_data = T)